Biogaran v. Lundbeck and the Challenger's Burden in Second Medical Use Cases

A second medical use patent is often seen as an easy target. The Milan section of the UPC Central Division has just shown that it is much harder to hit when the challenger is required to say exactly where the patent fails. On 9 October 2026, in Biogaran v. Lundbeck, the Court dismissed Biogaran's revocation action against Lundbeck's EP 4 223 296. The patent claims nalmefene, a molecule already known for the treatment of alcohol dependence, for reducing alcohol consumption in a specific group of patients. The Court maintained it in the amended form that Lundbeck had already filed with the EPO Opposition Division.1 Biogaran was ordered to bear the costs and to pay an interim award of EUR 100,000 within three weeks.2

The decision lays down three rules, each summarized in a headnote, and their reach goes well beyond nalmefene. First, the Court will not finish an objection of added matter that the claimant has left vague. Second, a claim limited to a group of patients does not need to state the dose, provided the skilled person can find a safe and effective one without undue effort. Third, when an obviousness case can only be built by combining more than ten documents, none of which points to the others, the case reflects hindsight, which the Court calls an ex post facto analysis.3

For holders of second medical use patents, the message is encouraging, with some reservations. The reason is a simple allocation of work: at every stage of the validity analysis, the Court asked the challenger to show precisely where the patent went wrong, and refused to do that work in the challenger's place. The result is all the more notable because, three years earlier, an EPO Board of Appeal had revoked an earlier patent of the same family, covering the same group of patients, for lack of inventive step.4

A patent with a difficult family history

EP 4 223 296 is a divisional patent, one of several descended from international application WO 2014/001427, filed on 27 June 2013. As construed by the Court, claim 1 covers nalmefene, taken as needed, to reduce alcohol consumption in alcohol-dependent patients who drink at a high-risk level (more than 60 g of pure alcohol a day for men, more than 40 g for women) and who are still drinking at that level one to two weeks after their first assessment.5 This observation period is the heart of the claim. It filters out the "early reducers", patients who drink less as soon as they enter a treatment setting. According to the patent's own figures, they made up more than half of the population studied.6

Biogaran brought its action on 14 November 2025 and raised the four classic grounds: added matter, insufficient disclosure, lack of novelty and lack of inventive step. In parallel, Lundbeck had asked the Paris Local Division for provisional measures based on this patent and on EP 3 345 604, but that application was withdrawn by agreement between the parties.7 At the oral hearing on 15 September 2026, Lundbeck turned its conditional application to amend into an unconditional one and deleted dependent claims 7 and 9. The text defended before the UPC thus became identical to the one filed with the Opposition Division after its preliminary opinion. The Court accepted this as a unilateral limitation of the patent.8 The value of the proceedings was set at EUR 2 million, as agreed by the parties, and the recoverable costs at half the applicable ceiling.9

Added matter: the challenger draws the map

Biogaran's main added-matter argument was structural. The original application, WO '427, lists 135 numbered embodiments, which can be combined in millions of ways. In Biogaran's view, the granted claim was therefore an arbitrary pick from that menu, unsupported by the original disclosure.10 Before looking at the substance, the Court set a rule of pleading. A party alleging added matter must compare the claim with the original disclosure, point by point, and identify exactly what has been extended. A long list of embodiments may raise questions of clarity or drafting style, but it does not in itself prove an extension. The Court added that, as an impartial body, it cannot build up an objection that the claimant has only stated in general terms. Doing so would amount to examining a ground of invalidity of its own motion (ex officio), contrary to the procedural fairness required by recital 5 of the Preamble to the Rules of Procedure.11

In practice, the Court was less strict than its headnote. Having found the attack insufficiently substantiated, it still compared the parent application with the patent, because the required level of detail remains uncertain and the Court of Appeal has not yet ruled on it. It openly admitted that this amounted, to a limited extent, to an ex officio assessment.12 It then found a direct basis for claim 1 in embodiments E24 to E26, which refer to one another and must be read together, and checked that they had been carried over unchanged into each application of the divisional chain. The words "and/or" in E24 expressly covered the case where both features are present. Claim 1 therefore did no more than select one of the options the disclosure already contained.13

The principle was therefore announced, not yet applied as a sanction. Its real weight lies in future cases. It targets a familiar argument against divisionals: that a parent application full of options cannot support any particular combination of them. After Milan, a challenger making that argument needs a map showing which feature is at stake, which passage it comes from, in which application of the chain the problem arises, and why the skilled person could not have derived the combination directly and unambiguously. A challenger who leaves the Court to draw that map should not expect it to do so a second time.

The dose the claim does not need

On sufficiency of disclosure, Biogaran's strongest point concerned the dose. Claim 1 mentions none, while the clinical trials described in the patent used only 18 mg. The claim, Biogaran argued, therefore also covered doses that would be ineffective or even toxic.14 The Court rejected the idea that a second medical use claim must, as a general rule, state the exact dosage in its independent claim. It relied on two points of EPO case law. Following G 2/21, which it found persuasive without treating it as binding, the therapeutic effect is part of what is claimed and must therefore be assessed under Article 83 EPC. Following T 0728/21, evidence published after the filing date cannot make credible an effect that was not credible at that date.15

The decisive step came next. Using nalmefene for alcohol dependence was already known. The contribution Lundbeck relied on was a stronger reduction in drinking within a narrowly defined group of patients. Once the invention lies in choosing the patients, the dose becomes a question of implementation. The description disclosed a range of 10 to 20 mg and an effective regimen of 18 mg. The closest prior art, a 2007 study by Karhuvaara and colleagues, already reported efficacy at 10 mg. And every skilled person knows that very low doses may not work and very high doses may be toxic. Finding a safe and effective regimen was therefore routine work. It was for Biogaran to prove that the invention would fail at doses other than 18 mg, and it had not done so.16

This is the passage most favorable to patentees, but also the one that depends most on the facts. It works because the patent's contribution lay in the choice of patients and not in the dose. A dosage regimen claim, where the dose itself is the invention, would call for a different analysis, and nothing in the decision suggests that the dose could then be left out of the claim. For drafters, the lesson is simple: put the dose ranges and an effective reference regimen in the description, and keep the claim at the level of the patient group. For challengers, the burden of proof matters more. An insufficiency attack based on the dose needs technical evidence that the effect is absent in part of the range. Pointing to what the claim does not say is not enough.

Thirteen documents and no pointer

The inventive-step attack started either from Karhuvaara or from a 2011 Lundbeck press release announcing the end of its Phase III program, each combined with up to twelve other documents.17 These ranged from the EMA guideline on medicinal products for alcohol dependence and studies on types of physiological dependence to cardiology trials with carvedilol and enalapril using run-in periods, and a cocaine-dependence study with a two-week lead-in period.18 The Court defined the objective technical problem as finding, among alcohol-dependent patients, a group in which nalmefene reduces drinking more markedly.19

The Court applied the test set by the Court of Appeal in Amgen v. Sanofi. Starting from a realistic point in the prior art, the skilled person must find more than an idea worth testing: the prior art must give a technically credible basis for reasonably predicting success. The Court distinguished a mere hope of success, which describes what the skilled person could try, from a reasonable expectation of success, which describes what he would actually do. It also referred to its own earlier ruling of 4 May 2026 on a second medical use.20 Karhuvaara contained no subgroup analysis and did not use the severity of dependence to select or allocate patients. None of the other documents mentioned nalmefene together with a selection criterion based on persistently high drinking, and several had nothing to do with alcohol. Relying on the Court of Appeal in Kodak v. Fujifilm, the Court held that finding the necessary elements scattered across different documents and contexts is not enough without a pointer, meaning a hint in the prior art that would lead the skilled person to combine them. Nor can the reason to combine come solely from the secondary document that the challenger has, with hindsight, chosen to add to the starting point.21

Only then did the Court address the mosaic. When an obviousness case can only be built by combining more than ten documents dealing with different technical questions, none of which provides the missing pointer or a reason to look at the others, the mosaic itself shows that the analysis is ex post facto.22 The number of documents is only a symptom. It confirms what the Court had already found document by document: none of them carried the pointer. An attack based on two documents that lacks a pointer fails for the same reason, simply less visibly.

Two Board decisions, one family

The strongest objection to an optimistic reading comes from the EPO itself. In T 0963/20, Board 3.3.07 revoked EP 3 138 564, an earlier patent of the same family. Its claim 1 covered nalmefene for alcohol-dependent patients whose high-risk drinking persisted after an observation period of one to two weeks. The Board accepted that the claim was new over Karhuvaara, but held it obvious. The 2010 EMA guideline recommended stratifying patients in clinical trials according to their level of dependence, and a skilled person applying that stratification to the Karhuvaara data would, in the Board's view, have arrived at the claimed group.23

Milan looked at the same combination and reached the opposite result. In its reading, the EMA guideline recommends stratification as a matter of trial methodology. It does not use the drinking risk level as a criterion for selecting patients, does not propose an observation period and does not suggest that nalmefene works better in the claimed group. The Court also refused to treat the Board's decision as res judicata. The two patents are legally distinct, and the Boards and the UPC belong to different legal systems: an EPO decision may persuade the UPC, but it cannot bind it.24

The EPO's own position has since moved. On 24 July 2026, in T 0894/25, the same Board dismissed the appeals of Tiefenbacher and Biogaran against the rejection of their oppositions to EP 3 345 604, the direct parent of the patent at issue, and found its claims inventive over the same documents. It explained why it departed from T 0963/20: a statistical analysis of the patent's trial data (D26), which it did not have in the earlier case, showed that stratifying patients by dependence scores does not single out the claimed group.25

One could read this divergence cynically, with the UPC as the patentees' forum and the EPO as the generics' forum. The better explanation lies in the test each body applied. In T 0963/20, the Board considered that a general suggestion to stratify patients by dependence was enough to make a specific selection criterion obvious. Milan asked a narrower question: would the skilled person have had a credible reason to expect that this particular criterion, persistent high drinking after an observation period, would predict a better response? That question helps patentees whose contribution is a specific selection rule backed by data showing a different effect in the selected group, which is exactly what the patent's response tables showed at the novelty stage.26 It gives no advantage to patentees in general. A claim to a patient group defined by a generic marker, without data on the differential effect, would fare no better in Milan than before the Board.

Questions left open

This is a first-instance decision, and an appeal may be lodged within two months of its service.27 The Court of Appeal might prefer the Board's reading of what a stratification guideline teaches, and the Milan panel itself openly awaits guidance on how specific an added-matter attack must be.28 The opposition against EP 4 223 296 is also still pending. The Opposition Division heard the parties on 25 August 2026, but no written decision appears in the Register yet, and any decision can be appealed. If the patent were revoked there, the revocation would apply in all designated states, whatever the outcome in Milan.29 Infringement was not decided, since the Paris application for provisional measures was withdrawn. Two questions remain open: whether the reasoning on dose extends to claims where the dosage regimen is itself the invention, and whether other divisions will adopt the "more than ten documents" rule of thumb.

Lessons for both sides

For holders of claims limited to a group of patients, the decision rewards choices made long before any litigation: selection criteria with real clinical meaning, backed by data showing that the effect differs in the claimed group, and a description that discloses the dose range and an effective regimen. It also shows the value of defending the same text before the UPC and in the opposition, which spared Lundbeck any debate about inconsistent claim sets. For generic companies, each ground now comes with its own burden. An added-matter attack must follow each feature through the divisional chain. An insufficiency attack on the dose must be backed by evidence. An obviousness attack must identify the document that provides the pointer and the basis for expecting success.

The contrast with T 0963/20 also matters when choosing where to litigate. Three decisions examined the same starting point and the same EMA guideline, and only the first found the claim obvious. For a challenger, the choice of forum therefore cannot be separated from the way the inventive-step case is built. A generic company facing a patient-group patent should first find the one document that carries the pointer, and only then decide whether to fight in Milan or before the Opposition Division.

• 1

  UPC Court of First Instance, Central Division (Milan section), 9 October 2026, UPC_CFI_1587/2025, Biogaran v. H. Lundbeck A/S, operative part and paras. 5 to 8.

• 2Ibid., paras. 276 to 279 and operative part.
• 3Ibid., headnotes 1 to 3.
• 4EPO Board of Appeal 3.3.07, T 0963/20, 7 November 2023, concerning EP 3 138 564.
• 5Biogaran v. Lundbeck, paras. 63 and 74 to 90.
• 6Ibid., paras. 191 to 198 and 253.
• 7Ibid., summary of facts and requests of the parties.
• 8Ibid., paras. 5 to 10 and 24; Rules 30.2, 49.2 and 50.2 RoP.
• 9Ibid., paras. 274 to 279.
• 10Ibid., paras. 104 to 105 and 123.
• 11Ibid., paras. 111 and 113 to 115.
• 12Ibid., paras. 116 to 118.
• 13Ibid., paras. 119 to 127.
• 14Ibid., paras. 157 to 165.
• 15Ibid., paras. 165, 170, 171 and 173 to 174, referring to G 2/21 of 23 March 2023, OJ EPO 2023, A85, Reasons, points 74 to 77, and to T 0728/21.
• 16Ibid., paras. 175 to 180.
• 17Ibid., paras. 226 to 230 and 236 to 237.
• 18Ibid., paras. 240 to 272.
• 19Ibid., para. 222.
• 20Ibid., paras. 221 and 243 to 245, citing UPC Court of Appeal, 25 November 2025, UPC_CoA_528/2024 and UPC_CoA_529/2024, Amgen v. Sanofi and Regeneron, and UPC Central Division (Milan section), 4 May 2026, UPC_CFI_552/2025, headnote 3.
• 21Ibid., paras. 233 to 234, 255 and 264 to 265, citing UPC Court of Appeal, 2 June 2026, UPC_CoA_312/2025 et al., Kodak v. Fujifilm, para. 183, and UPC Court of Appeal, 13 July 2026, UPC_CoA_473/2025 et al., Fujifilm v. Kodak, headnote 6.
• 22Ibid., para. 256 and headnote 3.
• 23T 0963/20, Reasons, point 2.3 (novelty) and points 3.1 to 3.4, in particular point 3.3.4 (inventive step).
• 24Biogaran v. Lundbeck, paras. 224 to 225 and 240 to 241.
• 25EPO Board of Appeal 3.3.07, T 0894/25, 24 July 2026, concerning EP 3 345 604, Reasons, points 6.3.3 to 6.3.5.
• 26Ibid., paras. 191 to 198.
• 27Article 73(1) UPCA; Rules 220.1(a) and 224.1(a) RoP.
• 28Biogaran v. Lundbeck, para. 116.
• 29Article 68 EPC; European Patent Register, EP 4 223 296 (application 23163177.1), consulted 18 October 2026.